WIP 30 - Prediction of Pediatric Non-Invasive Ventilation Failure Using Real World Data: A Retrospective Cohort Study

Poster Number: WIP 30
Publication Number: WIP 30.813

Background: Noninvasive ventilation (NIV) is an important strategy in the supportive care of moderate-to-severe respiratory illness in children. NIV can improve oxygenation and ventilation and is associated with fewer complications as compared to invasive mechanical ventilation. Early identification of patients at high risk of NIV failure is critical to ensure patients are transferred to appropriate levels of care and to avoid complications associated with delayed intubation and mechanical ventilation. In 2017, Duan et al. established a scoring system which predicts NIV failure in hypoxemic adult patients using heart rate, acidosis, consciousness (Glasgow Coma Scale), oxygenation, and respiratory rate (HACOR). Previous studies in pediatric literature have sought to describe factors associated with pediatric NIV failure but an accurate, validated prediction model or scoring system remains to be established in the pediatric population.

Objective: The objective of this study is to develop a model for noninvasive ventilation failure in pediatric patients one month to five years of age with undifferentiated acute respiratory distress using real world data in a setting where invasive mechanical ventilation is not available.

Design/Methods: The study was approved by the Colorado Multiple Institutional Review Board (COMIRB #23-1248) with waiver of consent. Data was collected from a prior randomized control trial that compared mortality in patients placed on continuous positive airway pressure (CPAP) versus control in a low-resource setting for which informed consent was obtained and documented by each patient’s legal guardian. For the present study, only the CPAP group will be analyzed (n = 1020). Baseline variables will be summarized using median (interquartile range) or n (%). A Kaplan Meier curve will be reported for time-to-death. Cox proportional hazard modeling will be used to build a model for time-to-death using vital signs and clinical characteristics. The concordance index will be reported. Data analysis is expected to be complete by December 2023.